When executing a job, outputs are stored in outputs/<output name> as a single file. This approach does not work very well for interactive editors since they may make many different outputs at different points in time and it would be nice to store the metadata in a single directory (along with the data) rather than writing all of them to result-types.json.